
SiMa.ai releases Palette 3.0 for running multiple AI models on one chip
The AMW Read
The release updates an edge AI infrastructure provider's software, but its performance claims lack test details and independent verification.
SiMa.ai releases Palette 3.0 for running multiple AI models on one chip
SiMa.ai has announced Palette 3.0, a software suite designed to run multiple AI models simultaneously on a single chip. The US physical AI semiconductor startup says the release builds on Palette Knit, its agent-based software environment announced in June. SiMa.ai claims the update can increase token processing throughput by up to 3.5 times in low-power environments. The report does not specify the comparison baseline or test conditions.
The release targets an infrastructure challenge in physical AI: coordinating several models on devices with limited power. SiMa.ai says the software improves processing for perception, depth, and reasoning models used in drones and robotics. It also adds open-source features for generative AI development. Sensor fusion and digital twin simulation capabilities, developed with partners, are intended to address integration and validation bottlenecks. The company did not name those partners in the report.
For builders, the practical question is whether concurrent models meet application requirements for throughput and power under realistic workloads. The company also says its existing agent-based environment can shorten development from months to days or hours, but the report provides no supporting test details. Investors assessing SiMa.ai's software and silicon strategy therefore have product claims to examine, while independent performance and development-time evidence remains unclear from this report.


